Outlier Patent Identification Based on Anomaly Detection- Taking the Field of UAV as an Example

We identify outlier patents from different perspectives to obtain more comprehensive identification results and assist in outlier innovation. Firstly, we use the BERT model to vectorize patent titles and abstracts to address the issue of polysemy; Then, we identify outlier patents through various anomaly detection algorithms and compare their recognition results; Finally, we obtain outlier patent topics through the BERTopic model. The experiment found that the outlier patent topics in the UAV field mainly focus on automatic charging, path planning, and 3D modeling.